Terms of use

The short version

• You describe a job; an agent does the reading and the drafting; you approve anything that leaves your company.
• What an agent does on your behalf is yours. You are responsible for it in the same way you are responsible for what an employee sends.
• We are honest about what is not built, on the page where you would meet it. Nothing here is charged for today.
• You can stop at any time, take every access grant away at Google, and ask for your data to be removed.

Who this is between

You, meaning the person signing up and the company whose workspace it is, and Force Labs, the people who make this product. If you are signing up on behalf of a company, you are saying you are allowed to.

A workspace has an Owner. The Owner decides who else is in it and what they may do, and the Owner is who we deal with about the workspace.

What you may use it for

Work you are allowed to do yourself. An agent acts with the access you give it, so the rule is simple: if you could not do it by hand with your own accounts, do not ask an agent to do it.

Specifically, do not use it to:

• Reach an account or a system you have not been given access to.
• Read a mailbox whose owner has not agreed to it. If the mailbox belongs to somebody else at your company, that is their decision to make, not yours.
• Get around a site's own terms, its rate limits, or a measure it uses to tell people from software.
• Send anything to anybody who has not asked to hear from you.
• Do anything illegal where you are, or where the site is.

You keep your own house in order: the accounts you connect, who is in your workspace, and what you approve. An approval is a decision; the product makes you make it, and then it is yours.

What we do, and what we do not promise

We run the product, keep the agent inside the permissions its definition grants, stop it before anything consequential and show you what it is about to do, and keep a record you can read afterwards.

We do not promise that an agent is right. A model reads pages and writes drafts, and both can be wrong: a page can be out of date, a reference can be misread, a draft can say something you would not have said. Every claim in an answer is checked against the pages the agent actually observed, which proves it saw what it says it saw and proves nothing about whether the page was telling the truth. Read the draft before you approve it.

Not built yetThere is no uptime commitment and no support response time. This is early software, parts of it are not switched on, and the product says which parts on the page where you would meet them. If something matters enough to need a guarantee, it does not have one yet.

Money

Nothing is charged today. There is no payment processor in this product, no card is stored anywhere in it, and every button on the pricing page opens a waitlist rather than a checkout. The prices shown are planned and are marked as planned.

When there is billing, it will be for what your agents use, meaning the model, the hosted browser and the messages, and it will be said before it starts, not after.

Anything an agent buys for you is not on our bill. That is your money going to a shop, and it never passes through us. Completing a payment is not built; an agent fills the basket and stops at your approval.

Your content, and ours

Yours stays yours

Your mail, your documents, the pages your agents read, the agents you build and what they produce. We claim no ownership of any of it.

What we may do with it

Run the product: hand the relevant part to the model, the browser or the messaging provider so the agent can do the job, and keep the record of what happened so you can check it. Nothing else.

Training

Your content is not used to train anybody's model. Model requests are made with data collection denied and zero-data-retention endpoints required where the gateway supports it.

Ours stays ours

The product itself, its design and its code.

Stopping

You can stop at any time. Take the access grants away at Google, remove the connected accounts, and sign out everywhere. Each of those works today and works immediately.

We can close an account that is being used for something in the list above, or that is costing other people the service. We would tell you why.

Not built yetDeleting an account and all of its data on request is not built. A request is recorded with a date and done by hand. The Privacy page says the same thing in the same words.

If something goes wrong

Tell us. There is a Feedback page in the product that reaches a person rather than a form that reaches nobody.

The legal part of this section, what we are liable for and what we are not, is the part most needing a lawyer, and no lawyer has read this document. It is deliberately left unwritten rather than filled with a paragraph copied from somewhere else, because a limitation of liability you do not understand protects nobody in the room.

Changes

Last written 18 October 2026. When these change in a way that matters, the account holder is told. A change that is only wording is not worth an email and will not get one.

Also

• Privacy
• Security: what stops an agent, and where it stops
• What is not live yet, the product’s own list

Last written 18 October 2026.